OpenAI retracts 3 of 722 AI math manuscripts over a sign error; $15M fund launched to help humans verify AI proofs
新智元 · wechat · 2026-11-10
OpenAI published 722 AI-generated math manuscripts on Oct 6; within a day, 3 were retracted over a single sign error (two more using the same construction were pulled too) and 14 revised. Only 42% of main results have Lean formalization. On Oct 15, the Simons Foundation announced a $15M commitment with OpenAI and Anthropic ($5M each, plus models, compute and Claude access) to help mathematicians verify, understand and extend AI outputs—responding to the AHM's harsh statement calling the dump a "display of power" and AGMAI's demand that labs fund human verification. Terence Tao has argued math is moving from proof scarcity to proof abundance, while Simons president David Spergel stressed the effort must be led by the math community. The piece also notes GPT-6 Astra scored 49/50 (98%) on TheoPhysEval after experts found errors in the reference answers themselves.
More from Research
- Dejan's LLM brand study probes 193k brand mentions for just $391.71 using Gemini — dejanseo · 2026-11-13
- Galaxea Co-founder Hang Zhao's CoRL Keynote: Split Robot Intelligence into System 2/1/0 by Timescale — YuXiang_IRVL · 2026-11-13
- A Summary of the Macy Conferences: The Birthplace of Cybernetics and Early AI Thought — ArtificialOther · 2026-11-13
- vismatch: unified API wrapping 50+ image matching models, 943 stars on GitHub — gabriberton · 2026-11-13
- Stanford's Speculative Self-Distillation cuts training tokens 38-69% with less forgetting — khademinori · 2026-11-13
- EEPO: Adaptive Sample-Then-Forget Rollouts Break Diversity Collapse in RLVR Training — burkov · 2026-11-13